Package 'fixes'

Description

S3 method that plots an es_result (from run_es()). It forwards arguments to plot_es().

Usage

## S3 method for class 'es_result'
autoplot(object, ci_level = 0.95, type = "ribbon", ...)

Arguments

Value

A ggplot object.

Examples

# res <- run_es(...)
# ggplot2::autoplot(res, ci_level = 0.95, type = "ribbon")

Description

Estimates average treatment effects on the treated (ATT) using either the Callaway-Sant'Anna (2021) or Borusyak-Jaravel-Spiess (2024) estimator, aggregated to a single summary effect ("simple"), per-cohort effects ("by_cohort"), or per calendar-time effects ("by_time").

Usage

calc_att(
  data,
  outcome,
  treatment = NULL,
  time,
  timing,
  fe = NULL,
  covariates = NULL,
  cluster = NULL,
  weights = NULL,
  interval = 1,
  time_transform = FALSE,
  unit = NULL,
  estimator = c("cs", "bjs"),
  aggregation = c("simple", "by_cohort", "by_time"),
  control_group = c("nevertreated", "notyettreated"),
  anticipation = 0L,
  conf.level = 0.95,
  vcov = "HC1",
  vcov_args = list()
)

Arguments

Details

This function complements run_es(): use run_es() when you want a full event-study curve (dynamic effects by relative time), and calc_att() when you want aggregated ATT estimates that collapse the time dimension.

Value

A data.frame of class "att_result" with columns:

group

Cohort or calendar time (NA for "simple").

estimate

ATT point estimate.

std.error

Standard error.

statistic

t-statistic (estimate / std.error).

p.value

Two-sided p-value (normal approximation).

conf_low_XX, conf_high_XX

CI bounds for each conf.level.

Attributes: aggregation, estimator, conf.level, N, N_units, N_treated, N_nevertreated, control_group (CS only), att_gt (CS raw ATT(g,t) table), tau_it (BJS unit-time effects table).

Aggregation formulas (CS estimator)

• simple: $\theta = \sum_g (n_g/n_{treated}) \cdot \overline{ATT(g,\cdot)}$ where $\overline{ATT(g,\cdot)}$ is the mean over post-treatment periods.

• by_cohort: $\theta(g) = \overline{ATT(g,\cdot)}$ per cohort.

• by_time: $\theta(t) = \sum_{g \le t} w(g,t) \cdot ATT(g,t)$ with $w(g,t) = n_g / \sum_{g' \le t} n_{g'}$.

Standard errors (BJS estimator)

BJS SEs are approximate (naive sample variance of unit-time effects). Cluster-robust SE for BJS aggregations is planned for v0.12.

See Also

run_es() for event-study (dynamic) estimates.

Description

Estimates the contamination weights $\omega^{\ell''}_{e,\ell}$ that decompose each TWFE event-study coefficient $\hat\mu_{\ell''}$ into a linear combination of cohort-specific ATTs (CATTs):

$\hat\mu_{\ell''} = \sum_{(e,\ell)} \omega^{\ell''}_{e,\ell} \cdot \widehat{CATT}_{e,\ell}$

(Sun and Abraham 2021, Equation 20).

The weights are obtained via the OVB auxiliary regression (Eq. 12): for each cohort-period CATT cell $(e, \ell)$, regress the cohort-specific indicator $1\{E_i = e\}\cdot 1\{t-E_i = \ell\}$ on all cohort-aggregated relative-time indicators $D^{\ell''}_{i,t}$ and two-way fixed effects. The resulting regression coefficient on $D^{\ell''}_{i,t}$ is $\omega^{\ell''}_{e,\ell}$.

Usage

compute_contamination_weights(
  data,
  time,
  timing,
  unit,
  fe = NULL,
  baseline = -1L
)

Arguments

Value

An object of class c("sa_contamination_weights", "data.frame") with one row per ⁠(catt_cohort, catt_period, twfe_period)⁠ triple, and columns:

catt_cohort

Cohort $e$ (first treatment period).

catt_period

Relative event time $\ell$ of the CATT.

twfe_period

Relative event time $\ell''$ of the TWFE coefficient being decomposed.

weight

Contamination weight $\omega^{\ell''}_{e,\ell}$.

is_own

Logical; TRUE when catt_period == twfe_period.

Attributes: baseline, cohorts, cohort_sizes, incl_periods.

Interpretation

• Own-period cell (catt_period == twfe_period): $\omega^{\ell}_{{e,\ell}}$ represents the weight the TWFE estimator places on $CATT_{e,\ell}$. Under the SA IW estimator these equal the cohort-size weights $n_e / \sum n_{e'}$.

• Cross-period cell (catt_period != twfe_period): Any non-zero weight indicates contamination: the TWFE coefficient $\hat\mu_{\ell''}$ also picks up treatment effects from period $\ell \ne \ell''$.

• Verification: the OVB identity (property iii) holds exactly, so $\hat\mu_{\ell''} = \sum_{(e,\ell)} \omega^{\ell''}_{e,\ell} \cdot \widehat{CATT}_{e,\ell}$ up to floating-point precision.

References

Sun, L. and Abraham, S. (2021). Estimating dynamic treatment effects in event studies with heterogeneous treatment effects. Journal of Econometrics, 225(2), 175–199.

See Also

plot_contamination_weights(), run_es()

Examples

## Not run: 
# Estimate contamination weights
cw <- compute_contamination_weights(
  data     = panel_data,
  time     = year,
  timing   = first_treat,
  unit     = id,
  fe       = ~ id + year,
  baseline = -1L
)
print(cw)
plot_contamination_weights(cw)

## End(Not run)

Description

Returns a single-row summary of model-level statistics from a run_did() result. Delegates to broom::glance.fixest() which provides nobs, r.squared, adj.r.squared, within.r.squared, AIC, BIC, and related statistics.

Usage

glance.did_result(x, ...)

Arguments

Value

A one-row data frame of model-level statistics.

Examples

## Not run: 
res <- run_did(df, outcome = y, treatment = D, fe = ~ id + year)
broom::glance(res)

## End(Not run)

Description

Robust inference and sensitivity analysis for event-study / DiD designs following Rambachan and Roth (2023). Instead of assuming parallel trends holds exactly, it asks how large a violation of parallel trends would have to be before the causal conclusion changes, returning confidence sets for a post-treatment effect under a sequence of restrictions on the possible difference in trends.

Two restriction families are supported:

• "relative_magnitude" ($\Delta^{RM}(\bar M)$): post-treatment violations are at most $\bar M$ times the largest pre-treatment violation (Rambachan & Roth 2023, Section 2.4.1).

• "smoothness" ($\Delta^{SD}(M)$): the difference in trends deviates from linearity by at most $M$ per period (Section 2.4.3).

Inference uses the Andrews-Roth-Pakes (ARP) conditional moment-inequality test (Section 3.2.1), which is uniformly valid and recommended for general restriction sets.

Usage

honest_sensitivity(
  object = NULL,
  type = c("relative_magnitude", "smoothness"),
  Mvec = NULL,
  l_vec = NULL,
  alpha = 0.05,
  gridPoints = 1000L,
  betahat = NULL,
  sigma = NULL,
  numPrePeriods = NULL,
  numPostPeriods = NULL
)

Arguments

Value

A data.frame of class "honest_result" with one row per restriction value plus the original (parallel-trends) confidence interval, with columns M, lb, ub, method, and type. The breakdown value (largest restriction at which the robust CI still excludes 0) is stored in attr(., "breakdown").

References

Rambachan, A. and Roth, J. (2023). A More Credible Approach to Parallel Trends. Review of Economic Studies, 90(5), 2555-2591.

See Also

run_es(), plot_honest()

Description

Visualises the full cohort-by-period ATT(g,t) matrix stored in the att_gt attribute of an es_result object produced by run_es(estimator = "cs"). Two display styles are available:

• "heatmap": a tile plot with calendar time $t$ on the x-axis and cohort $g$ on the y-axis, colour-filled by the point estimate. Cells whose pointwise confidence interval excludes zero are marked with a filled dot; cells that are simultaneously significant (when bootstrap data are available) are additionally marked with an open diamond.

• "facet": one panel per cohort showing ATT(g,t) over calendar time $t$ with a pointwise confidence ribbon, mirroring the style of plot_es(). A lighter simultaneous CI ribbon is overlaid when bootstrap data are available.

Both types draw a vertical dashed line at $t = g$ (treatment onset) for each cohort.

Usage

plot_att_gt(
  x,
  type = c("heatmap", "facet"),
  ci_level = 0.95,
  zero_line = TRUE,
  theme = c("bw", "minimal", "classic"),
  color = "#B25D91FF",
  fill = "#B25D91FF",
  alpha = 0.2
)

## S3 method for class 'att_gt_result'
autoplot(object, ...)

Arguments

Value

A ggplot2::ggplot() object.

Bootstrap annotation

When attr(x, "bootstrap") is present (i.e., run_es() was called with bootstrap = TRUE), both plot types add simultaneous inference overlays sourced from the (g,t)-level bootstrap object.

See Also

plot_es(), run_es()

Examples

## Not run: 
cs_result <- run_es(data = mydata, outcome = y, time = year,
                    timing = g, unit = id, fe = ~id + year,
                    staggered = TRUE, estimator = "cs")
plot_att_gt(cs_result)
plot_att_gt(cs_result, type = "facet")

## End(Not run)

Description

Creates a ggplot2 tile heatmap of the contamination weights returned by compute_contamination_weights().

Each cell at position (twfe_period, catt_label) shows the weight $\omega^{\ell''}_{e,\ell}$: how much of $CATT_{e,\ell}$ leaks into the TWFE coefficient $\hat\mu_{\ell''}$.

• Diagonal cells (catt_period == twfe_period): own-period weights (sum across cohorts should be $\approx 1$).

• Off-diagonal cells: cross-period contamination (ideally close to zero under treatment effect homogeneity).

Usage

plot_contamination_weights(
  x,
  limit_abs = NULL,
  midpoint = 0,
  low = "#2166AC",
  mid = "white",
  high = "#B2182B",
  theme = c("bw", "minimal", "classic"),
  show_values = FALSE,
  value_digits = 2L
)

Arguments

Value

A ggplot2::ggplot() object.

See Also

compute_contamination_weights()

Description

Plot event-study results with ribbons or error bars

Usage

plot_es(
  data,
  ci_level = 0.95,
  type = "ribbon",
  vline_val = 0,
  vline_color = "#000",
  hline_val = 0,
  hline_color = "#000",
  linewidth = 1,
  pointsize = 2,
  alpha = 0.2,
  barwidth = 0.2,
  color = "#B25D91FF",
  fill = "#B25D91FF",
  theme_style = "bw",
  show_simultaneous = FALSE
)

Arguments

Value

A ggplot object.

Examples

# Assuming `res <- run_es(...)`
# p <- plot_es(res, ci_level = 0.95, type = "ribbon")
# print(p)

Description

Creates an interactive plotly visualization of event study results with hover-over displays showing coefficients, confidence intervals, and other details.

Usage

plot_es_interactive(
  data,
  ci_level = 0.95,
  vline_val = 0,
  hline_val = 0,
  vline_color = "#000",
  hline_color = "#000",
  color = "#B25D91FF",
  fill = "#B25D91FF",
  alpha = 0.2,
  linewidth = 2,
  markersize = 8,
  show_ribbon = TRUE,
  show_simultaneous = FALSE,
  height = NULL,
  width = NULL
)

Arguments

Details

The hover tooltip displays:

• Relative time to treatment

• Point estimate (coefficient)

• Confidence interval bounds

• Standard error

• P-value

• Simultaneous CI bounds (when show_simultaneous = TRUE)

Value

A plotly object that can be displayed interactively.

Examples

## Not run: 
# Assuming res <- run_es(...)
plot_es_interactive(res)
plot_es_interactive(res, ci_level = 0.99, show_ribbon = FALSE)
plot_es_interactive(res, show_simultaneous = TRUE)

## End(Not run)

Description

Visualises the output of honest_sensitivity(): robust confidence intervals for the target effect under progressively weaker parallel-trends restrictions (increasing $M$ or $\bar M$), alongside the original confidence interval that assumes parallel trends holds exactly. This is the "top-down" sensitivity plot of Rambachan and Roth (2023).

Usage

plot_honest(x, ...)

## S3 method for class 'honest_result'
autoplot(object, ...)

Arguments

Value

A ggplot object.

See Also

honest_sensitivity()

Description

Estimates a classic difference-in-differences model of the form outcome ~ D_it | fe using fixest::feols().

There are two ways to supply the treatment indicator:

Option A — pre-built D_it (maximum flexibility):

df$D <- as.integer(df$treated & df$year >= 2006)
run_did(df, outcome = y, treatment = D, fe = ~ id + year)

Option B — timing-based construction (convenience; consistent with run_es() and calc_att()):

run_did(df, outcome = y, treatment = treated, time = year, timing = 2006,
        fe = ~ id + year)

Here treatment is a binary group indicator (1 = treated unit, 0 = control), time is the calendar-time variable, and timing is the scalar treatment onset period. Internally D_it = treatment * (time >= timing) is constructed automatically. For staggered-adoption settings use calc_att(); for dynamic event-study estimates use run_es().

Usage

run_did(
  data,
  outcome,
  treatment,
  timing = NULL,
  fe = NULL,
  unit = NULL,
  time = NULL,
  covariates = NULL,
  cluster = NULL,
  weights = NULL,
  conf.level = 0.95,
  vcov = "HC1",
  vcov_args = list()
)

Arguments

Value

A did_result object (named list) with elements:

estimates

Data frame with the treatment coefficient: term, estimate, std.error, statistic, p.value, and conf_low_XX/conf_high_XX for each conf.level entry.

model

The underlying fixest model object.

Attributes: call, formula_str, outcome, treatment, timing, fe, vcov_type, cluster_vars, conf.level, N, N_units, N_treated, unit, time.

Examples

## Not run: 
# Option A: pre-built D_it
df$D <- as.integer(df$treated & df$year >= 2006)
res <- run_did(df, outcome = y, treatment = D, fe = ~ id + year)

# Option B: timing-based construction
res <- run_did(df, outcome = y, treatment = treated, time = year,
               timing = 2006, fe = ~ id + year)

# Cluster-robust SE
res <- run_did(df, outcome = y, treatment = D, fe = ~ id + year,
               cluster = ~ id)

print(res)
broom::tidy(res)
broom::glance(res)
# modelsummary::modelsummary(res)

## End(Not run)

Description

Runs an event study regression on panel data, supporting both classic (universal timing) and staggered (unit-varying timing via sunab). The function builds the design (lead/lag factor or sunab), estimates with fixest::feols(), and returns a tidy table with metadata.

Usage

run_es(
  data,
  outcome,
  treatment = NULL,
  time,
  timing,
  fe = NULL,
  lead_range = NULL,
  lag_range = NULL,
  covariates = NULL,
  cluster = NULL,
  weights = NULL,
  baseline = -1L,
  interval = 1,
  time_transform = FALSE,
  unit = NULL,
  staggered = FALSE,
  method = c("classic", "sunab"),
  estimator = c("twfe", "cs", "sa", "bjs", "twm", "flex"),
  control_group = c("nevertreated", "notyettreated"),
  anticipation = 0L,
  conf.level = 0.95,
  vcov = "HC1",
  vcov_args = list(),
  bootstrap = FALSE,
  B = 999L,
  alpha = 0.05,
  boot_seed = NULL,
  group = NULL,
  trends = FALSE
)

Arguments

Value

A data.frame of class "es_result" with columns:

• term, estimate, std.error, statistic, p.value

• conf_low_XX, conf_high_XX (for each requested conf.level)

• relative_time (integer; 0 = event), is_baseline (logical; marks the reference period)

Attributes include: lead_range, lag_range, baseline, interval, call, model_formula, conf.level, N, N_units, N_treated, N_nevertreated, fe, vcov_type, cluster_vars, staggered, sunab_used.

Key Features

• One-step event study: specify outcome, treatment, time, timing, fixed effects, and (optionally) covariates.

• Switch between Classic (factor expansion) and Staggered-SAFE (method = "sunab").

• Flexible clustering, weights, and VCOV choices (e.g., ⁠vcov = "HC1" | "HC3" | "CR2" | "iid" ...⁠).

• Automatic lead/lag window detection and customizable baseline period.

• Returns an "es_result" object compatible with print() and autoplot().

Description

Returns a tidy data frame of model coefficients from a run_did() result. Delegates to broom::tidy.fixest() on the underlying fixest model so that all regressors (treatment and covariates) appear in the output — the format expected by modelsummary::modelsummary().

Usage

## S3 method for class 'did_result'
tidy(x, conf.int = FALSE, conf.level = 0.95, ...)

Arguments

Value

A data frame with columns term, estimate, std.error, statistic, p.value (and optionally conf.low, conf.high).

Examples

## Not run: 
res <- run_did(df, outcome = y, treatment = D, fe = ~ id + year)
broom::tidy(res)
broom::tidy(res, conf.int = TRUE)

## End(Not run)